Orders, rosters and replies, done before service
Phase 2 for the same three-venue restaurant group. The second brain knew how every dish is made. Now AI workflows draft the orders, the rosters and the replies, and the owner gets the numbers every Monday.
01 At a glance
The short version
Phase 1 gave the group a second brain. Phase 2 puts it to work on the two biggest costs in any restaurant: food and wages.
A second brain, three months in
Questions to the owner down from 35 to 12 a week. Prep lists drafted daily for all three kitchens. The same dish at every venue.
Waste, wages and slow replies
Orders written late at night from memory. Rosters copied from last week. Reviews and function enquiries waiting days.
Four AI workflows
Stock and ordering, rosters, reviews and enquiries, and the Monday numbers. Each one with a clear person who signs off.
$10,030 a month by month 6
78 team hours back a month, less food in the bin and tighter rosters. Delivered on top of the $5,240 from phase 1.
The shift
Phase 1: the AI answers questions. Phase 2: the AI does the work between services. It drafts the order, the roster and the reply. A chef or manager still signs off.
02 The new pain
The recipes were sorted.
The margins weren't.
Once the questions stopped, the next problem was easy to see. The venues were busy, but money leaked out the back door.
Orders written at 11pm
Chefs counted the cool room after close and ordered from memory. Too much of some things, not enough of others.
About $1,200 of food binned a week
Across three venues. Nobody linked what was booked to what was ordered and prepped.
Rosters built on Sunday night
Managers copied last week's roster. Quiet nights were overstaffed, and overtime only showed up on payday.
Reviews and function enquiries waited
Replies took days. One enquiry for a 40-person function went cold over a long weekend.
The owner saw the numbers a month late
Wage and food cost came from the accountant at month end. Too late to fix anything that month.
03 What we built
The second brain,
now with hands
AI operations (AI that does routine work, not just answers). Four workflows read the recipes, par levels (how much stock to keep) and rules from the second brain. They work through the booking system, the till and the rostering app the group already used.
Three things changed
A stocktake, a week of bookings, a new review. Each one starts a workflow without anyone remembering to.
Par levels, roster rules and reply wording the owner signed off once. Anything new waits for a person.
Every Monday the owner sees what the workflows did, and what they couldn't.
Human review stayed in
Chefs approve every order. Managers approve every roster and every reply to an unhappy guest. The AI never changes a recipe or an allergen card, and allergy questions still go to a person.
04 The four workflows
Four workflows,
one lighter bin
Each one has a clear trigger and a clear person who signs off. We added one a month and ran each with full approval for two weeks first.
| Workflow | Starts when | What it does | Who approves |
|---|---|---|---|
| Stock and ordering | The closing chef enters the stocktake | Drafts each venue's order from par levels, bookings and last week's sales. Flags supplier price jumps | The head chef or sous chef at that venue |
| Rosters | Next week's bookings land | Drafts the roster to the covers expected. Flags overtime and missed breaks | The venue manager |
| Reviews and enquiries | A review or function enquiry arrives | Drafts a reply in the venue's voice. Sends function menus and deposit details | The manager. Complaints go to a person |
| Monday numbers | Every Monday, 6am | Sales, wage cost, food cost and waste, by venue | Read by the owner, nothing is sent |
Allergies and upset guests
An allergy question, a complaint, a staff issue. These go straight to a person, the same shift.
Bookings predict the week
140 covers booked is a better guide than last Monday's gut feel. The order follows the bookings.
05 One week at the bistro
An order, a roster,
a review
Here's one week at the busiest venue. Work that used to happen at 11pm or on Sunday night now happens in minutes.
Stocktake after close
The closing chef counts the cool room on a tablet. Twenty minutes, then home.
Order drafted
Monday's order is drafted from par levels and 140 covers booked. Lamb shoulder is cut back after Saturday's leftovers.
Chef approves
The head chef changes one line and approves. Orders reach the suppliers by 7:30am.
Roster drafted
Next Tuesday is light on bookings, so one floor shift is cut. The manager approves it in five minutes.
Review answered
A 3-star review about slow mains. A reply is drafted, the manager softens one line and sends it. The head chef gets a note.
In the owner's numbers
Waste down, wage cost down, review answered. All counted on one page.
Before phase 2
The order would have been written at 11pm from memory. The Tuesday roster would have been last week's. And the review would have sat unanswered until someone had a quiet hour.
06 The owner's Monday
The numbers, ready
before the first coffee
Every Monday at 6am the owner gets one page. It replaces waiting a month for the accountant.
Across three venues. Bistro up 6%, café flat.
Down from 34%. The wine bar is still at 36%, worth a look.
Waste $690, mostly bread and dairy at the café.
Who signs off on what
| The AI can do on its own | A person approves first |
|---|---|
| Draft each venue's order | Sending it. The chef approves every order |
| Draft next week's roster | Publishing it to staff |
| Reply to 4 and 5-star reviews from approved wording | Any reply to a review of 3 stars or less |
| Send function menus and deposit details | Any discount, refund or change to a deposit |
The rule we built everything on
The AI can draft and count. Anything that spends money, changes someone's shift or answers an unhappy guest waits for a person.
07 Delivered outcome
Delivered operational impact
at month 6
Each operational improvement is tracked directly from Monday venue reporting.
| Area | Before phase 2 | Delivered by month 6 |
|---|---|---|
| Stock orders | Written at 11pm from memory | Drafted by 6am, approved in minutes |
| Food waste | About $1,200 a week | About $700 a week |
| Rosters | Copied on Sunday night | Drafted from bookings, overtime flagged first |
| Reviews and function enquiries | Days to reply, some never | Same day, in the venue's voice |
| The owner's numbers | A month late, from the accountant | One page, Monday 6am |
The 90-day rollout
Stocktakes on a tablet, draft orders live. Two weeks with every order checked line by line.
Draft rosters started. Review and enquiry replies went live. The Monday numbers began.
All four ran on their own within the rules. The team checks a sample each week.
Verified operational impact
These numbers are drawn directly from POS data, inventory logs and Monday reports across the first six months of operation.
08 The numbers
Delivered value
worth $10,030 a month
Measured across three commercial drivers: team hours returned, food not binned, and rostered hours trimmed.
Orders that follow the bookings, not a guess.
Mostly quiet early-week shifts.
8 manager, 7 chef and 3 owner hours by month 6.
Month by month
| Measure | Month 4 | Month 5 | Month 6 |
|---|---|---|---|
| Team hours back a week | 6 | 13 | 18 |
| Value of time back a month | $1,840 | $3,980 | $5,590 |
| Food waste saved a month | $650 | $1,520 | $2,170 |
| Wages saved a month | $0 | $1,210 | $2,270 |
| Total a month | $2,490 | $6,710 | $10,030 |
A year at the month 6 rate. With phase 1, that's about $15,270 a month, or $183,200 a year.